The Business Automation Guide

The best automation candidates share three characteristics: high frequency, low judgement variance, and a measurable downstream consequence. Rare, high-judgement work is the last thing you should automate, not the first.

Score each candidate on annual value and build effort. Start in the high-value, low-effort quadrant even if it feels unambitious, the early credibility is worth more than the sophistication.

Deliberately defer the high-effort, high-value work until the organisation trusts the pattern. Trust is the real constraint on automation programmes, not capability.

Every automation should answer four questions before a line of it is built: what happens when the input is malformed, when confidence is low, when an external system is unavailable, and when the output is wrong and discovered later.

Systems that cannot answer these questions are prototypes, regardless of how well they perform in a demonstration.

Place a human gate wherever the cost of an error exceeds the cost of the review. That calculation changes over time: gates that are essential in month one are often removable by month six once error rates are known.

Review gates should be fast. If approving an item takes longer than doing it manually, the design has failed.

Automations decay. Upstream systems change, business rules move, and edge cases accumulate. Assign an owner and a review cadence to every automation, and retire the ones that no longer earn their place.

Track four numbers per automation: volume processed, exception rate, human review time, and the commercial metric it exists to influence. Without the fourth, you are measuring activity.